Anthropic has launched a brand new model of its mid-size Sonnet mannequin, protecting tempo with the corporate’s four-month replace cycle. In a post announcing the new model, Anthropic emphasised enhancements in coding, instruction-following, and pc use.
Sonnet 4.6 would be the default mannequin for Free and Professional plan customers.
The beta launch of Sonnet 4.6 will embody a context window of 1 million tokens, twice the dimensions of the biggest window beforehand out there for Sonnet. Anthropic described the brand new context window as “sufficient to carry complete codebases, prolonged contracts, or dozens of analysis papers in a single request.”
The discharge comes simply two weeks after the launch of Opus 4.6, with an up to date Haiku mannequin prone to comply with within the coming weeks.
The launch comes with a brand new set of document benchmark scores, together with OS World for pc use and SWE-Bench for software program engineering. However maybe probably the most spectacular is its 60.4% rating on ARC-AGI-2, meant to measure expertise particular to human intelligence. The rating places Sonnet 4.6 above most comparable fashions, though it nonetheless trails fashions like Opus 4.6, Gemini 3 Deep Assume, and one refined model of GPT 5.2.
